There is a strange assumption inside many AI strategies.
The company buys a tool. The company announces a training programme. The company asks people to share use cases. Then it waits for its cleverest employees to come forward and explain how they have made themselves faster.
Why would they?
If I have found a way to do three hours of work in forty minutes, I have learned something valuable. I might share it because I am generous. I might share it because I believe in the team. I might share it because the company makes that feel safe.
Or I might reasonably wonder whether my reward for sharing will be more work, a tighter target, fewer people on my team, or a new colleague who can now do the part of my job that made me useful.
That is not an adoption problem. It is an incentive problem wearing a cheerful lanyard.
AI is already entering through the side door
The official AI programme is rarely the whole story.
Recent US evidence finds adoption at three separate levels: firms, business functions, and individual tasks. Workers can use AI where their employer has not formally adopted it; employers can adopt it without it becoming part of day-to-day worker practice. Among firms using AI, use is still usually narrow rather than company-wide. NBER
The pattern appears elsewhere. In an ILO study of Polish workplaces, more people reported using generative AI than said their employer had officially introduced it. The report describes technology entering work through private accounts and personal devices—faster than policy, training, or management can keep up. ILO
This should not surprise anyone who has watched Excel, Google, Slack, or a decent template arrive in an organisation. People adopt tools around the official process when the official process cannot help them meet a deadline.
What is different now is that the private shortcut can change the shape of someone’s work.
It can make a person faster at writing, research, coding, analysis, preparation, planning, and translation. It can make them look unusually organised. It can let them attempt work that previously required another function. The personal advantage is not just a trick. It is a new rate of learning.
Silence can be a rational response
Leaders sometimes treat hidden AI use as a compliance failure. It can be. Sensitive data in an unapproved tool is a real risk.
But the silence also tells you something about the company.
People hide a useful practice when they do not know whether it will be rewarded, confiscated, punished, or misunderstood. They hide it when sharing means writing a mini business case in the hope of receiving a meeting about a meeting. They hide it when the people who set the policy do not understand the work well enough to distinguish a safe draft from an unsafe decision.
Then the organisation gets the worst of both worlds. It loses visibility into where value is being found, and workers lose a safe route to turn a personal breakthrough into shared capability.
Everyone calls this a knowledge-management problem. It is also a trust problem.
The question that gives the game away
Ask a team: “What happens to someone who shares the best AI workflow they have built?”
If the answer is vague, you have found the reason the workflow is still private.
The company does not need a compulsory prompt museum. Nobody wants to upload their half-finished thoughts into a shared folder called Innovation and wait for them to die there.
It needs a fair exchange.
Give people protected time to turn a useful pattern into something another person can use. Credit the person who found it. Let them retain some ownership of the next version. Separate experimentation from a demand to immediately raise output targets. Make the risk boundary clear enough that people know what can be shared safely.
Most important: do not treat the discovery of a better way to work as evidence that the old workload was imaginary.
If a person becomes more capable, the company has a choice. It can turn that capability into a higher quota. Or it can make the capability travel.
One produces a quiet arms race.
The other produces a company that learns.